Proximity-Aware Classroom Monitoring: Enhancing Action Recognition with Contextual Intelligence

This paper proposes a Context-Aware Action Recognition System for monitoring in classrooms, which combines spatio-temporal modeling with context awareness to boost robustness as well as interpretability. As compared to traditional frame-wise models, the system combines deep learning-based action recognition with spatial and temporal processing to yield contextual information, e.g., the recognition of actions near specific objects. A hybrid proximity analysis combining bounding box overlap with IoU-based thresholds provides contextual information, e.g., the recognition of actions near objects. Experimental results on multiple datasets verify its effectiveness in spatial relationship encoding. Additionally, the system provides real-time monitoring, anomaly detection, and statistical analysis, thus promoting enhanced improved classroom management.